Industrial internet of things data trusted transmission method based on edge computing

Abstract

The present application relates to the technical field of industrial internet of things communication security and edge computing, and more particularly to an industrial internet of things data trusted transmission method based on edge computing, which is applied to an edge node device and comprises: acquiring multi-source time series collection data and alternating voltage waveform data; extracting a zero-crossing time point to generate a hardware physical timestamp combined with a key, and calculating a local sliding variance; outputting a data drift prediction gradient through an autoregressive prediction network layer; constructing a penalty coefficient dynamic optimization model, inputting a false alarm cost function and a bandwidth constraint boundary into the optimization model, iteratively outputting optimal physical correlation sensitivity parameters, and fusing the local sliding variance and a static threshold to generate a dynamic scheduling physical threshold; and comparing the data drift prediction gradient with the physical threshold, and generating pre-emptive transmission trigger signaling when the gradient exceeds the threshold, so as to transfer transparent data payload containing the physical timestamp and the collection data to a network card sending queue to perform overruling transmission, and realize precursor data trusted pre-emptive backhaul.

Description

Technical Field

[0001] This invention relates to the fields of industrial IoT communication security and edge computing technology, and in particular to a reliable data transmission method for industrial IoT based on edge computing. Background Technology

[0002] In industrial IoT edge computing scenarios, predictive maintenance of heavy rotating machinery or precision manufacturing equipment requires extremely high real-time data transmission security. Traditional industrial edge node data transmission mostly adopts static threshold alarm mechanisms based on the current absolute amplitude or post-event network status feedback. Because static feedback mechanisms cannot perceive the early dynamic evolution trend of time-series data, they are prone to causing communication queue congestion in high-concurrency network environments in factories. This results in critical precursor data that is in the very early stages of a fault, with amplitude not exceeding the limit but showing a microscopic deterioration trend, being forcibly dropped or delayed in transmission by the network queue. Summary of the Invention

[0003] To overcome the above shortcomings, this invention provides a reliable data transmission method for the Industrial Internet of Things based on edge computing. It aims to improve the problem that traditional edge node transmission mostly uses fixed static thresholds, which cannot adaptively predict data deterioration trends, resulting in weak precursory data being discarded by the congested network.

[0004] This invention provides the following technical solution: a trusted data transmission method for industrial IoT based on edge computing, applied to edge node devices, comprising the following steps: S1. Acquire multi-source time series data of the target process point and acquire AC voltage waveform data input to the power supply terminal of the edge node device; S2. Extract the continuous zero-crossing time points of the AC voltage waveform data, generate hardware-level physical timestamps by combining them with the node key, and calculate the local sliding variance of the AC voltage waveform data. S3. Input the multi-source time series data into a preset autoregressive prediction network layer and output the data drift prediction gradient of the multi-source time series data. S4. Construct a dynamic optimization model for the penalty coefficient, input the industrial fault underreporting cost function and the physical link bandwidth congestion constraint boundary into the dynamic optimization model for the penalty coefficient for iterative calculation, and output the optimal physical correlation sensitivity parameter. S5. The optimal physical correlation sensitivity parameter, the local sliding variance and the benchmark static threshold are fused together to generate the optimal dynamic scheduling physical threshold. S6. Compare the data drift prediction gradient with the optimal dynamic scheduling physical threshold. When the data drift prediction gradient is greater than or equal to the optimal dynamic scheduling physical threshold, generate a preemptive transmission trigger signaling. S7. By using the preemptive transmission trigger signaling, the transparent transmission data payload containing the hardware-level physical timestamp and the multi-source time series acquisition data is transferred to the network card transmission queue, and a cross-level network transmission action is performed.

[0005] By adopting the above technical solution, a dynamic optimization model is constructed to generate dynamic scheduling physical thresholds and combine them with predicted gradients to execute preemptive transmission, thereby realizing the cross-level transparent transmission of fault precursor data. This improves the problem that traditional edge node transmission mostly uses fixed static thresholds, which cannot adaptively predict the data deterioration trend, resulting in weak precursor data being discarded by the congested network.

[0006] The present invention has the following beneficial effects: 1. In this invention, a dynamic optimization model is constructed to generate a dynamic scheduling physical threshold and combined with a predicted gradient to execute preemptive transmission, thereby realizing the cross-level transparent transmission of fault precursor data. This improves the problem that traditional edge node transmission mostly uses fixed static thresholds, which cannot adaptively predict the data deterioration trend, resulting in weak precursor data being discarded by the congested network.

[0007] 2. In this invention, hardware-level physical timestamps are generated by extracting continuous zero-crossing time points of AC voltage waveform data and combining them with node keys. Then, transparent transmission data payloads are encapsulated, thereby improving the problem that traditional IoT authentication mostly uses pure software timestamps. Since software clocks are easily monitored and tampered with, this can lead to the malicious replay of historical normal data.

[0008] 3. In this invention, by inputting the collected data into a preset autoregressive prediction network layer, the data drift prediction gradient of the collected data is output in advance, thereby improving the problem that traditional industrial monitoring mostly uses the current absolute amplitude comparison, which lacks the advance calculation of the deterioration trend of time series data, resulting in the missed detection of very early weak fault precursors of equipment.

[0009] 4. In this invention, by inputting the fault underreporting cost function and the bandwidth constraint boundary into the optimization model to iteratively output the sensitivity parameter, and then integrating the power grid variance to generate the optimal dynamic scheduling threshold, the problem of network congestion and collapse caused by the inaccurate threshold adjustment due to the failure to comprehensively consider the risk of underreporting in traditional network scheduling, which mostly adopts one-way index evaluation, is improved. Attached Figure Description

[0010] Figure 1 This is a flowchart of the industrial IoT data trusted transmission method based on edge computing proposed in this invention; Figure 2 This is a data acquisition and physical timestamp generation architecture diagram for the industrial IoT data trusted transmission method based on edge computing proposed in this invention; Figure 3This is a detailed architecture diagram of the model prediction and dynamic optimization of the industrial IoT data trusted transmission method based on edge computing proposed in this invention. Figure 4 This is a detailed architecture diagram of the scheduling physical threshold generation and level triggering of the industrial IoT data trusted transmission method based on edge computing proposed in this invention; Figure 5 This is a detailed architecture diagram of the x-preemptive transparent network transmission method for trusted data transmission in the Industrial Internet of Things based on edge computing proposed in this invention. Detailed Implementation

[0011] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] Example 1: In a first embodiment of the present invention, the present invention provides a reliable data transmission method for the Industrial Internet of Things based on edge computing, such as... Figures 1-2 As shown, it includes the following steps: S1. Acquire multi-source time series data of the target process point and acquire AC voltage waveform data input to the power supply terminal of the edge node device; Furthermore, in S1, multi-source time-series acquisition data of the target process point is obtained, and AC voltage waveform data input to the power supply terminal of the edge node device is acquired, including: Continuous monitoring of mechanical operation physical parameters is achieved by deploying a multimodal sensor array at the target process points; The mechanical operation physical parameters are converted into discrete digital signal sequences by the first analog-to-digital converter, and the discrete digital signal sequences are written into the memory receiving buffer as multi-source time series acquisition data. A parallel connection is formed between the hardware step-down transformer circuit and the physical AC power supply terminals of the edge node device; The analog voltage changes of the AC power supply physical terminals are acquired in real time by the second analog-to-digital converter to generate AC voltage waveform data. By forcibly triggering synchronous sampling interrupt instructions for the first and second analog-to-digital converters using a low-level hardware timer, a hardware-level synchronization alignment mechanism is established between multi-source time series acquisition data and AC voltage waveform data in the time dimension.

[0013] Specifically, in step S1, multi-source time series acquisition data of the target process point is obtained and AC voltage waveform data input from the power supply terminal of the edge node device is obtained.

[0014] A multimodal sensor array is deployed at the target process point to continuously monitor the physical parameters of mechanical operation. These parameters flow via physical cables into the analog input channel of a first analog-to-digital converter (ADC). The ADC performs discretization sampling, converting the parameters into a discrete digital signal sequence. This discrete digital signal sequence, as multi-source time-series data, is directionally written into the memory receive buffer allocated to the edge node device, awaiting subsequent computation by the computing module.

[0015] The hardware step-down transformer circuit inside the edge node device forms a parallel connection topology with the physical AC power supply terminals of the edge node device. The analog voltage change signal at the physical AC power supply terminals is stepped down and conditioned before being input to the sampling port of the second analog-to-digital converter (ADC). The second ADC acquires the analog voltage change signal in real time and outputs discrete AC voltage waveform data.

[0016] The underlying hardware timer forcibly triggers a synchronous sampling interrupt instruction according to the set clock frequency. The synchronous sampling interrupt instruction is directly sent to the external trigger pins of the first and second analog-to-digital converters via a hardwired bus. When the first and second analog-to-digital converters capture the rising edge of the synchronous sampling interrupt instruction, they forcibly perform data latching, establishing a hardware-level synchronization alignment mechanism between the multi-source time series acquisition data and the AC voltage waveform data in the time dimension.

[0017] The discrete sampling mathematical model for the hardware-level synchronization alignment mechanism is defined as follows.

[0018] Set the synchronous sampling clock period issued by the underlying hardware timer to be The physical parameters of mechanical operation in the continuous time domain are defined as follows: The analog voltage change signal in the continuous time domain is defined as... In the first Discrete multi-source time series data extracted at each sampling interruption trigger time. Discrete AC voltage waveform data The relationship is as follows: ; ; in This represents the variable indicating the sequence number of discrete sampling points. Represents a continuous physical time variable. This represents the fixed analog-to-digital conversion delay time at the circuit level for the first analog-to-digital converter. This indicates the fixed analog-to-digital conversion delay time at the circuit level for the second analog-to-digital converter. It relies on a clock reference generated by the same underlying hardware timer. The driving sampling action achieves absolute time anchoring between multi-source time series data and AC voltage waveform data at the physical level, completely eliminating random time difference jitter caused by the multi-threaded polling mechanism at the operating system software level. The two aligned discrete data streams flow from their respective analog-to-digital converter outputs to the subsequent feature extraction and model prediction modules.

[0019] S2. Extract the continuous zero-crossing time points of the AC voltage waveform data, combine them with the node key to generate a hardware-level physical timestamp, and calculate the local sliding variance of the AC voltage waveform data. Furthermore, in S2, continuous zero-crossing time points of the AC voltage waveform data are extracted and combined with the node key to generate hardware-level physical timestamps, including: Acquire the transient voltage value and peak voltage value of AC voltage waveform data within the current sampling period; The voltage ratio parameter is obtained by dividing the transient voltage value by the peak voltage value. The phase offset is generated by multiplying the preset reference zero-crossing time interval with the voltage ratio parameter. Extract the target zero-crossing time point that is closest to the current acquisition action among consecutive zero-crossing time points; The initial authentication string is generated by performing bit-concatenation logic operations on the target zero-crossing time point, phase offset, and node key; The initial authentication string is input into the secure hash algorithm function configured in the internal hardware encryption engine of the edge node device for one-way hash operation, and the fixed-length hash value output by the hash operation is used as the hardware-level physical timestamp.

[0020] In S2, the local sliding variance of the AC voltage waveform data is calculated, including: Calculate the time span between adjacent zero-crossing time points in a series of zero-crossing time points to construct a zero-crossing time interval sequence; The average value of the baseline interval is obtained by summing all the values ​​in the zero-crossing time interval sequence and dividing by the total number of values ​​in the sequence. Calculate the deviation of each value in the zero-crossing time interval sequence from the average value of the reference interval; The arithmetic mean of the deviation is obtained by multiplying it by itself and then taking the square, which gives the local sliding variance that characterizes the intensity of transient jitter of mechanical load in industrial power grid.

[0021] Specifically, in step S2, the edge node device uses a hardware zero-crossing comparator circuit containing a Schmitt trigger to extract consecutive zero-crossing time points of the AC voltage waveform data to construct a time reference. The Schmitt trigger has a built-in hysteresis voltage threshold to eliminate electromagnetic high-frequency oscillation interference near the zero-level crossover point of the AC voltage waveform data. The microcontroller acquires the transient voltage value and peak voltage value of the AC voltage waveform data within the current sampling period. The arithmetic logic unit of the microcontroller divides the transient voltage value and the peak voltage value to obtain a voltage ratio parameter. The arithmetic logic unit performs a product operation based on the preset reference zero-crossing time interval and the voltage ratio parameter to generate a phase offset. The underlying mathematical model for calculating the phase offset is defined as follows: ; ; in This represents the voltage ratio parameter. This represents the transient voltage value within the current sampling period. This represents the peak voltage value of the AC voltage waveform data. This indicates the phase offset. This indicates the preset reference zero-crossing time interval.

[0022] The microcontroller traverses the data sequence of consecutive zero-crossing time points. It extracts the time value closest to the current acquisition action from the consecutive zero-crossing time points as the target zero-crossing time point. The microcontroller's bit manipulation unit performs bit concatenation logic operations on the 64-bit target zero-crossing time point, the 32-bit phase offset, and the 128-bit node key in a strictly fixed order from high to low bits. The bit reconstruction model of the 224-bit initial authentication string generated by the bit concatenation logic operation is as follows: ; in This represents the initial authentication string, which is 224 bits long. This indicates the zero-crossing time point of the 64-bit target. This indicates a bitwise concatenation logical operator. This represents the 32-bit phase offset. This represents the 128-bit node key allocated internally by the edge node device.

[0023] The initial authentication string is input to the edge node device's internal hardware encryption engine via the internal data bus. The hardware encryption engine, configured with the national standard SM3 secure hash algorithm, performs a one-way hash operation on the initial authentication string. The 256-bit fixed-length hash value output by the secure hash algorithm engine is used as a hardware-level physical timestamp and written into a memory structure awaiting network packet encapsulation. This timestamp generation logic, based on the physical characteristics of the power supply, replaces the conventional operating system software clock reading operation, preventing network-level time replay tampering.

[0024] In step S2, the microcontroller synchronously calculates the local sliding variance of the AC voltage waveform data according to the set time window. The microcontroller calculates the time span between adjacent zero-crossing points in a series of zero-crossing time points to construct a discrete sequence of zero-crossing time intervals. The arithmetic logic unit sums all the values ​​in the zero-crossing time interval sequence and divides by the total number of values ​​in the sequence to obtain the average value of the reference interval.

[0025] The arithmetic logic unit (ALU) calculates the deviation of each value in the zero-crossing time interval sequence from the average of the reference intervals. The deviation is squared using a hardware multiplier. The ALU then calculates the arithmetic mean of all the squared results. The output of this arithmetic mean represents the local sliding variance. This local sliding variance is output from the ALU and cached in memory for subsequent use by the optimization model. Variations in the variance data characterize the transient jitter intensity of the mechanical load in the industrial power grid. The computational model for this physical feature extraction process is as follows: ; ; ; in and These represent the th time points in the sequence of consecutive zero-crossing time points. The and the first Zero-crossing time sampling points. It represents the time span between adjacent zero-crossing points. It represents the total number of values ​​contained in the zero-crossing time interval sequence. This represents the average value of the baseline interval. This represents the local sliding variance. By calculating the discreteness characteristics of the physical interval sequence, the abstract power grid load change is converted into a quantifiable variance value, which is then input into the dynamic optimization algorithm.

[0026] like Figure 3 As shown, S3, input the multi-source time series data into the preset autoregressive prediction network layer, and output the data drift prediction gradient of the multi-source time series data; Furthermore, in S3, the multi-source time series data is input into a pre-defined autoregressive prediction network layer, and the data drift prediction gradient of the multi-source time series data is output, including: The input nodes of the pre-defined autoregressive prediction network layer receive the actual observed physical amplitude of multi-source time series data at the current sampling time. The actual observed physical amplitude is weighted and summed with the historical predicted physical amplitude at the previous moment to output the target predicted physical amplitude at the next moment. Subtracting the actual observed physical amplitude from the predicted physical amplitude of the target yields the physical amplitude evolution bias. Divide the physical amplitude evolution deviation by the preset sampling time step constant to obtain the data drift prediction gradient.

[0027] Specifically, in step S3, multi-source time-series data flows from the memory receiving buffer of the edge node device into a preset autoregressive prediction model. The input node of the preset autoregressive prediction model receives the actual observed physical amplitude of the multi-source time-series data at the current sampling time. The preset autoregressive prediction model reads the historical predicted physical amplitude from the previous time step from its internal register. The arithmetic unit inside the preset autoregressive prediction model multiplies the actual observed physical amplitude and the historical predicted physical amplitude from the previous time step by a set weighting coefficient to perform a weighted summation calculation. The sum of the observation weighting constant assigned to the current actual observed physical amplitude and the historical weighting constant assigned to the historical predicted physical amplitude is strictly equal to 1, and the specific value is determined by online adaptive fitting of historical fault samples using the least squares method. The output data of the weighted summation action is stored in the buffer queue as the target predicted physical amplitude for the next time step. The internal underlying algorithm mapping and weight constraint model of this weighted evaluation stage are as follows: ; ; in This represents the predicted physical magnitude of the target at the next moment. This represents the actual observed physical amplitude at the current sampling moment. This represents the historical predicted physical amplitude at the previous moment. This represents the observation weighting constant assigned to the current actual observed physical amplitude. This represents the historical weighting constant assigned to the historical predicted physical amplitude.

[0028] The arithmetic operation unit retrieves the predicted physical amplitude of the target and performs an arithmetic subtraction operation with the actual observed physical amplitude. The result of the subtraction operation represents the deviation in the evolution of the physical amplitude. The corresponding numerical processing model is as follows: ; in This represents the deviation in the evolution of the physical amplitude. The arithmetic unit retrieves a preset sampling time step constant from the memory configuration space. The arithmetic unit then divides the deviation in the physical amplitude evolution by the preset sampling time step constant using a hardware divider. This division directly generates the data drift prediction gradient. This calculation process is equivalent to solving a discrete rate of change model of the sensor's physical amplitude within the current time window.

[0029] ; in This represents the gradient for predicting data drift. This represents the preset sampling time step constant. The calculated data drift prediction gradient is sent to the bus via the output pin of the preset autoregressive prediction network layer. The bus transmits the data drift prediction gradient and latches it in the designated address space of the comparison register, awaiting subsequent state comparison calls. This completes the closed loop of time series data processing from physical amplitude input to rate of change gradient output.

[0030] like Figure 3 As shown in Figure S4, construct a dynamic optimization model for the penalty coefficient, input the industrial fault underreporting cost function and the physical link bandwidth congestion constraint boundary into the dynamic optimization model for the penalty coefficient for iterative calculation, and output the optimal physical correlation sensitivity parameter. Furthermore, in S4, a dynamic optimization model for the penalty coefficient is constructed. The industrial fault underreporting cost function and the physical link bandwidth congestion constraint boundary are input into the dynamic optimization model for the penalty coefficient, including: Set a physical correlation sensitivity coefficient to characterize the penalty for threshold reduction and a network congestion dynamic pricing factor; The cost function for industrial failure underreporting is generated by multiplying the basic downtime economic loss constant with a negative exponential function constructed from the physical correlation sensitivity coefficient. The burst throughput prediction value is obtained by dividing the preset network throughput adjustment weight by the physical correlation sensitivity coefficient. Subtracting the physical network interface card's maximum communication bandwidth from the burst throughput prediction value generates the physical link bandwidth congestion constraint boundary. Multiplying the network congestion dynamic pricing factor by the physical link bandwidth congestion constraint boundary yields the Lagrange penalty term. The cost function for industrial fault underreporting is added to the Lagrange penalty term to construct the Lagrange dual optimization equation, which is then used as a dynamic optimization model for the penalty coefficient.

[0031] In S4, iterative calculations are performed to output the optimal physical correlation sensitivity parameters, including: Obtain the first descent iteration step size and the second descent iteration step size; Based on the partial derivative expression of the physical association sensitivity coefficient of the dynamic optimization model with respect to the first descent iteration step size and penalty coefficient, the physical association sensitivity coefficient for the next iteration round is calculated and updated. Based on the second descent iteration step size and the physical link bandwidth congestion constraint boundary, calculate and update the network congestion dynamic pricing factor for the next iteration round. Compare the absolute value of the difference between the physical correlation sensitivity coefficients of two adjacent iterations; When the absolute value of the difference is less than the built-in lower limit of the convergence error, the subgradient iteration calculation is stopped, and the physical correlation sensitivity coefficient in the converged state is output as the optimal physical correlation sensitivity parameter.

[0032] Specifically, in step S4, the arithmetic logic unit of the edge node device retrieves preset initial value data from the system memory to start the iterative solver. The arithmetic logic unit sets a physical correlation sensitivity coefficient representing the threshold reduction penalty in a register. The arithmetic logic unit sets a network congestion dynamic pricing factor in another separate register.

[0033] The arithmetic logic unit (ALU) reads the basic downtime economic loss constant stored in non-volatile memory. The ALU then calls the floating-point unit to construct a negative exponential function by taking the negative value of the physical correlation sensitivity coefficient and using it as the exponent of the natural logarithm. The floating-point unit multiplies the basic downtime economic loss constant with this negative exponential function to generate the industrial failure underreporting cost function. The corresponding mathematical calculation model is expressed as follows: ; in This represents the cost function for missed industrial fault reports. This represents the basic downtime economic loss constant. This represents the physical correlation sensitivity coefficient.

[0034] The arithmetic logic unit (ALU) reads the preset network throughput adjustment weights. The divider divides the network throughput adjustment weights by the physical correlation sensitivity coefficient to obtain the burst throughput prediction value. The ALU obtains the physical upper limit of the physical network interface card's (NIC) maximum communication bandwidth recorded in the NIC firmware. The ALU subtracts the physical upper limit of the NIC's maximum communication bandwidth from the burst throughput prediction value through a subtraction operation. The output data of this subtraction operation constitutes the physical link bandwidth congestion constraint boundary. The corresponding spatial constraint equation is defined as follows: ; in This indicates the physical link bandwidth congestion constraint boundary. This indicates the preset network throughput adjustment weight. This indicates the physical upper limit of the communication bandwidth of the physical network card.

[0035] The multiplier multiplies the network congestion dynamic pricing factor with the previously generated physical link bandwidth congestion constraint boundary to obtain the Lagrange penalty term. The arithmetic logic unit adds the industrial fault underreporting cost function and the Lagrange penalty term to form the Lagrange dual optimization equation. The Lagrange dual optimization equation is loaded into the solver's instruction queue as a dynamic optimization model with penalty coefficients. The merged equation is expressed as follows: ; in This represents the Lagrange dual optimization equation. This represents the dynamic pricing factor for network congestion.

[0036] After the model is built, iterative computation begins. The microcontroller sets the variable for the current iteration round as follows: The initial values ​​of the physical correlation sensitivity coefficient and the network congestion dynamic pricing factor are forcibly reset to zero. The arithmetic logic unit reads the first and second descent iteration step sizes from the configuration register. The arithmetic logic unit performs partial derivative calculations on the Lagrange dual optimization equation with respect to the physical correlation sensitivity coefficient and performs updates based on the first descent iteration step size. For the inequality bandwidth constraint, the arithmetic logic unit performs multiplier updates based on the second descent iteration step size and forces a superimposed non-negative projection truncation operation to maintain the mathematical validity of the convex optimization space. The underlying iterative update formula and partial derivative expansion are as follows: ; ; ; in This represents the partial derivative expansion of the penalty coefficient dynamic optimization model with respect to the physical correlation sensitivity coefficient. This represents the basic downtime economic loss constant. This indicates the network throughput adjustment weight. and They represent the first Rounds and Number The sensitivity coefficient of physical association of rounds. This indicates the step size of the first descent iteration. and They represent the first Rounds and Number Dynamic pricing factor for network congestion in each round. This indicates the step size of the second descent iteration. Indicates the first The physical link bandwidth congestion constraint boundary calculated in each round. The operator represents extracting the maximum value from two input values ​​to force the iteratively updated network congestion dynamic pricing factor onto a non-negative real space.

[0037] The digital comparator reads the physical correlation sensitivity coefficients output from two consecutive iterations and calculates the absolute value of the difference. The comparator then determines whether the absolute value of the difference is less than a set constant-level built-in convergence error lower limit and simultaneously checks whether the variable in the current iteration has reached the set maximum iteration limit. When the absolute value of the difference is less than the built-in convergence error lower limit or the variable in the current iteration has reached the maximum iteration limit, the microcontroller sends an interrupt signal to the solver to stop the iterative calculation, ensuring the real-time response capability of the control loop. The physical correlation sensitivity coefficient of the latest iteration is locked and output as the optimal physical correlation sensitivity parameter to the buffer data bus.

[0038] ; in This represents the absolute value of the difference between the physical correlation sensitivity coefficients of two adjacent iterations. The optimal physical correlation sensitivity parameter after the convergence condition is met becomes the direct input variable for subsequent threshold generation logic.

[0039] like Figure 4 As shown, S5, the optimal physical correlation sensitivity parameter, local sliding variance and benchmark static threshold are fused together to generate the optimal dynamic scheduling physical threshold; Furthermore, in S5, the optimal physical association sensitivity parameter, local sliding variance, and baseline static threshold are fused together to generate the optimal dynamic scheduling physical threshold, including: The optimal physical correlation sensitivity parameter is multiplied by the local sliding variance to output the power grid disturbance penalty correction amount; The baseline static threshold is read from the non-volatile memory, and the grid disturbance penalty correction is subtracted from the baseline static threshold to calculate the optimal dynamic scheduling physical threshold.

[0040] Specifically, in step S5, the edge node device performs dynamic reconstruction of the threshold based on the convergence parameters and feature parameters output by the preceding calculation. The arithmetic logic unit reads the optimal physical association sensitivity parameters output by the aforementioned solver from the cached data bus. The arithmetic logic unit retrieves the local sliding variance extracted within the time window from the internal random access memory.

[0041] The microcontroller's hardware multiplier multiplies the optimal physical correlation sensitivity parameter with the local sliding variance. The multiplier directly outputs a grid disturbance penalty correction, characterizing the impact of physical grid fluctuations on communication scheduling. This grid disturbance penalty correction is temporarily stored as an intermediate variable in the microcontroller's internal general-purpose data register. The digital model of the aforementioned multiplication operation is described below: ; in This represents the amount of power grid disturbance penalty correction generated. This represents the optimal physical correlation sensitivity parameter. This represents the local sliding variance.

[0042] The microcontroller initiates a data addressing request to the onboard non-volatile memory via the serial peripheral interface bus. The non-volatile memory responds to the read command and sends back the built-in reference static threshold to the microcontroller. The microcontroller's internal arithmetic logic unit reads the grid disturbance penalty correction temporarily stored in the general-purpose data register. The arithmetic logic unit performs an arithmetic subtraction operation, subtracting the grid disturbance penalty correction from the reference static threshold. The data stream output by the subtraction operation is the optimal dynamic scheduling physical threshold generated for the current physical operating conditions. The underlying formula for calculating the optimal dynamic scheduling physical threshold is as follows: ; in This represents the optimal dynamic scheduling physical threshold. This represents the baseline static threshold read from non-volatile memory.

[0043] After the optimal dynamic scheduling physical threshold is generated, it is pushed to the address space of the dedicated threshold register of the hardware logic comparison circuit via the internal instruction bus. The data is latched in this dedicated threshold register address space to serve as the dynamic reference value for subsequent comparison operations. The entire calculation process completes a closed loop of the fixed threshold adjustment mechanism based on the actual electromagnetic disturbance state.

[0044] like Figure 4 As shown, S6 compares the data drift prediction gradient with the optimal dynamic scheduling physical threshold. When the data drift prediction gradient is greater than or equal to the optimal dynamic scheduling physical threshold, a preemptive transmission trigger signaling is generated. Furthermore, in S6, the data drift prediction gradient is compared with the optimal dynamic scheduling physical threshold. When the data drift prediction gradient is greater than or equal to the optimal dynamic scheduling physical threshold, a preemptive transmission trigger signaling is generated, including: The hardware logic comparison circuit that inputs the data drift prediction gradient to the edge node device includes a positive digital comparison pin; The optimal dynamic scheduling physical threshold is input to the inverted digital comparison pin contained in the hardware logic comparison circuit; When the input value of the positive digital comparator pin is greater than or equal to the input value of the negative digital comparator pin, a continuous high-level pulse is output to the network card interrupt controller on the edge node device motherboard through the system data bus, and the high-level pulse is used as a preemptive transmission trigger signal.

[0045] Specifically, in step S6, the threshold determination logic within the edge node device is completely offloaded to the physical hardware layer for execution to eliminate interrupt response latency in the software system. The system bus controller routes the data drift prediction gradient output from the preceding calculation to the hardware logic comparison circuit of the edge node device. The data drift prediction gradient is input via the data bus to the positive digital comparison pin included in the hardware logic comparison circuit.

[0046] The optimal dynamic scheduling physical threshold, temporarily stored in the address space of the dedicated threshold register, is retrieved synchronously. This optimal dynamic scheduling physical threshold, as a dynamically changing reference signal, is input to the inverted digital compare pin of the hardware logic compare circuit.

[0047] The hardware logic comparator circuit performs a hardware-level state comparison of the input values ​​of the positive and negative digital comparator pins. The discrete truth-value judgment model for this hardware logic state comparison is expressed as follows: ; in This indicates the level state of the output pin of the hardware logic comparison circuit. This indicates the data drift prediction gradient input to the positive phase digital comparator pin. This represents the optimal dynamic scheduling physical threshold for input to the inverting digital compare pin. This indicates a continuous high-level pulse state. This indicates the default low-level hold state.

[0048] When the input value of the positive digital comparator pin is greater than or equal to the input value of the inverting digital comparator pin, the hardware logic comparator circuit deployed within the field-programmable gate array (FPGA) generates a nanosecond-level response. The gate array inside the hardware logic comparator circuit undergoes a state flip, and the output terminal breaks away from the default low-level holding state and generates a continuous high-level pulse.

[0049] This high-level pulse bypasses the CPU's interrupt vector table polling and is directly output to the network interface card (NIC) interrupt controller on the edge node device's motherboard via a separately configured system data bus. The high-level pulse's arrival at the NIC interrupt controller's hardware trigger endpoint is explicitly defined as a preemptive transmission trigger signaling. The NIC interrupt controller performs an edge-triggered latch operation on the preemptive transmission trigger signaling, preparing to activate the next stage of physical network transmission timing. This completes the closed-loop data link from data characteristic determination to pure physical control signal conversion.

[0050] like Figure 5 As shown, S7, through preemptive transmission trigger signaling, transfers the transparent transmission data payload containing hardware-level physical timestamps and multi-source time series data to the network card transmission queue, and performs a cross-level network transmission action. Furthermore, in S7, through preemptive transmission triggering signaling, the transparent transmission data payload, containing hardware-level physical timestamps and multi-source time-series data, is transferred to the network interface card's (NIC) transmission queue, executing cross-level network transmission actions, including: The preemptive transmission trigger signal is captured by the external interrupt pin of the network card interrupt controller, and the regular polling process of the underlying transmission control protocol being executed by the operating system is forcibly suspended. Activate the direct memory access channel and directly concatenate the hardware-level physical timestamp and multi-source time series data in the system memory space into a transparent transmission data payload that does not require secondary encapsulation by the application layer protocol. By using the active Direct Memory Access Channel, the transparent transmission data payload is forcibly and continuously copied to the hardware transmit buffer of the network card's media independent interface, and the hardware transmit buffer is used as the network card's transmit queue. Locate the byte offset of the Differentiated Services Control field in the message header structure layout, and rewrite the Differentiated Services Control field as the highest transmission priority identifier; By locking the exclusive right to use the transmission channel of the physical network interface device and bypassing the congestion control logic of the operating system's transport layer, the network data packet containing the transparent transmission data payload and the highest transmission priority identifier is directly sent to the external physical communication link, thus completing the cross-level network transmission action.

[0051] Specifically, in step S7, the hardware-level preemption control logic takes over the system's data transmission rights. The external interrupt pin of the network card interrupt controller captures the preemptive transmission trigger signaling transmitted upstream on the bus. This trigger signaling, as the highest priority non-maskable interrupt signal, interrupts the current execution flow of the microprocessor. The microprocessor responds to the interrupt instruction, forcibly suspending the underlying transmission control protocol regular polling process being executed by the operating system kernel.

[0052] The microprocessor then activates the direct memory access channel on the motherboard bus. Within the designated system memory space, hardware-level physical timestamps and multi-source time-series data are merged into memory blocks starting at specific addresses. This continuous memory mapping operation directly concatenates the data to generate a transparent transmission payload. This payload eliminates the overhead of additional application-layer protocol encapsulation. The direct concatenation parsing model at the data layer is expressed as follows: ; in This represents the transparent data payload generated in the system memory space. This represents the hardware-level physical timestamp calculated from the zero-crossing check. The byte stream concatenation operator represents a contiguous physical memory address space. This indicates multi-source time series data collection.

[0053] In its active state, the Direct Memory Access Channel (DMI) acquires control of the system bus. The DMI bypasses microprocessor register data transfer and continuously copies the transparent transmission payload from system memory to the hardware transmit buffer of the network interface card's (NIC) medium-independent interface. To avoid data read / write conflicts, the DMI applies a hardware mutex to the hardware transmit buffer during the copying process. Upon completion of the copy, the DMI triggers a transmission completion interrupt, releases the hardware mutex, and configures the hardware transmit buffer as a NIC transmit queue. The transparent transmission payload is carried over the User Datagram Protocol (UDP) to bypass the retransmission and connection establishment overhead of the Transmission Control Protocol (TCP).

[0054] The internal hardware state machine of the physical network interface device parses the layout of the Internet Protocol version 4 (IP4) packet header structure in the network interface card's (NIC) transmission queue. The hardware state machine locates the absolute byte offset of the Type of Service (Type of Service) field in the IPC header. The hardware state machine writes the Accelerated Forwarding Differential Service Code Point (CFS) with a decimal value of 46 to the Type of Service field, completing the rewriting assignment of the highest transmission priority identifier. The data rewriting model at the byte offset position is defined as follows: ; in This represents the header data structure of a network data packet. This indicates the absolute byte offset position of the Differentiated Services Control field in the message header data structure. This represents the highest transmission priority identifier constant pre-allocated by the system.

[0055] The physical network interface device triggers its built-in hardware mutex mechanism to lock exclusive access to the transmission channel. The network packet transmission process forcibly bypasses the operating system's transport layer congestion control logic at the hardware level, discarding the standard slow start backoff window mechanism. The physical layer transceiver driver circuit directly modulates and converts the network packet, containing the transparent transmission data payload and the highest transmission priority identifier, into a physical level signal. This physical level signal is then emitted through the port to the external physical communication link, thus completing the closed loop of a cross-level network transmission action independent of system software scheduling.

[0056] Example 2: This application focuses on high-frequency real-time monitoring of the spindle operation status of high-precision CNC machine tools in large manufacturing plants. In this Industrial Internet of Things (IIoT) scenario, edge node devices need to continuously collect time-series data on machine tool vibration and temperature and transmit it back to the cloud for predictive maintenance. The current technical challenges in this scenario are: First, severe electromagnetic interference and network eavesdropping in industrial settings can easily lead to the tampering of software timestamps based on the operating system, causing historical normal data to be maliciously replayed to mask the true fault state, resulting in unreliable data sources. Second, when multiple heavy machines operate concurrently in the plant, causing micro-vibrations in the local AC power grid, conventional fixed alarm thresholds cannot be adaptively adjusted, easily leading to data transmission storms across the network due to background noise or missing true, weak early signs of faults. Finally, when machine tools show very early signs of physical deterioration, the plant's wireless network is often in a state of high concurrency congestion. Traditional operating system transmission layer congestion control logic will forcibly slow down, queue, or discard such critical early warning data, causing the control center to miss the optimal intervention time and resulting in irreversible equipment damage. To address the aforementioned problems, this invention provides a reliable data transmission method for the Industrial Internet of Things (IIoT) based on edge computing, such as... Figure 1 As shown. The specific implementation process of this method is as follows: Edge node devices acquire multi-source time-series data and AC voltage waveform data. The microcontroller extracts continuous zero-crossing timestamps from the AC voltage waveform data, concatenates them with the node key to generate a hardware-level physical timestamp, and calculates the local sliding variance of the waveform data. The acquired data flows into the autoregressive prediction network layer, where the computation unit parses the time-series features and outputs the data drift prediction gradient. The arithmetic logic unit iterates through a dynamic optimization model using the industrial fault underreporting cost function and physical link bandwidth congestion constraints as boundary values, outputting the optimal physical correlation sensitivity parameter. This parameter, along with the local sliding variance and a baseline static threshold, is weighted and fused by the arithmetic logic unit to generate the optimal dynamic scheduling physical threshold. The hardware comparison circuit performs a level comparison between the data drift prediction gradient and this physical threshold. When the prediction gradient reaches or exceeds the physical threshold, the output pin level flips, generating a preemptive transmission trigger signal. The direct memory access channel responds to this signal, transferring the transparent transmission payload containing the physical timestamp and acquired data to the underlying network interface card's transmit queue, bypassing the operating system's conventional transmission control stack to directly execute a cross-level network transmission action.

[0057] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A trusted data transmission method for the Industrial Internet of Things (IIoT) based on edge computing, applied to edge node devices, characterized in that: Includes the following steps: S1. Acquire multi-source time series data of the target process point and acquire AC voltage waveform data input to the power supply terminal of the edge node device; S2. Extract the continuous zero-crossing time points of the AC voltage waveform data, generate hardware-level physical timestamps by combining them with the node key, and calculate the local sliding variance of the AC voltage waveform data. S3. Input the multi-source time series data into a preset autoregressive prediction network layer and output the data drift prediction gradient of the multi-source time series data. S4. Construct a dynamic optimization model for the penalty coefficient, input the industrial fault underreporting cost function and the physical link bandwidth congestion constraint boundary into the dynamic optimization model for the penalty coefficient for iterative calculation, and output the optimal physical correlation sensitivity parameter. S5. The optimal physical correlation sensitivity parameter, the local sliding variance and the benchmark static threshold are fused together to generate the optimal dynamic scheduling physical threshold. S6. Compare the data drift prediction gradient with the optimal dynamic scheduling physical threshold. When the data drift prediction gradient is greater than or equal to the optimal dynamic scheduling physical threshold, generate a preemptive transmission trigger signaling. S7. By using the preemptive transmission trigger signaling, the transparent transmission data payload containing the hardware-level physical timestamp and the multi-source time series acquisition data is transferred to the network card transmission queue, and a cross-level network transmission action is performed.

2. The industrial IoT data trusted transmission method based on edge computing according to claim 1, characterized in that, In S1, acquiring multi-source time-series data of the target process point and acquiring AC voltage waveform data input to the power supply terminal of the edge node device includes: The physical parameters of mechanical operation are continuously monitored by a multimodal sensor array deployed at the target process point. The mechanical operation physical parameters are converted into discrete digital signal sequences by a first analog-to-digital converter, and the discrete digital signal sequences are written into the memory receiving buffer as the multi-source time series acquisition data. The hardware step-down transformer circuit is connected in parallel with the physical AC power supply terminals of the edge node device. The analog voltage changes of the AC power supply physical terminals are acquired in real time by a second analog-to-digital converter to generate the AC voltage waveform data. By forcibly triggering a synchronous sampling interrupt instruction for the first analog-to-digital converter and the second analog-to-digital converter through a low-level hardware timer, a hardware-level synchronization alignment mechanism is established between the multi-source time series acquisition data and the AC voltage waveform data in the time dimension.

3. The industrial IoT data trusted transmission method based on edge computing according to claim 1, characterized in that, In S2, the step of extracting the continuous zero-crossing time points of the AC voltage waveform data and generating a hardware-level physical timestamp by combining the node key includes: Obtain the transient voltage value and peak voltage value of the AC voltage waveform data within the current sampling period; The voltage ratio parameter is obtained by dividing the transient voltage value by the peak voltage value. The phase offset is generated by multiplying the preset reference zero-crossing time interval with the voltage ratio parameter. Extract the target zero-crossing time point that is closest to the current acquisition action from the continuous zero-crossing time points; The target zero-crossing time point, the phase offset, and the node key are combined using bit-concatenation logic to generate an initial authentication string; The initial authentication string is input into the secure hash algorithm function configured in the internal hardware encryption engine of the edge node device for one-way hash operation, and the fixed-length hash value output by the hash operation is used as the hardware-level physical timestamp.

4. The industrial IoT data trusted transmission method based on edge computing according to claim 1, characterized in that, In S2, calculating the local sliding variance of the AC voltage waveform data includes: Calculate the time span between adjacent zero-crossing time points in the continuous zero-crossing time points to construct a zero-crossing time interval sequence; The average value of the baseline interval is obtained by summing all the values ​​in the zero-crossing time interval sequence and dividing by the total number of values ​​in the sequence. Calculate the deviation of each value in the zero-crossing time interval sequence from the average value of the reference interval; The arithmetic mean of the deviation is obtained by multiplying it by itself and then taking the square, which gives the local sliding variance that characterizes the transient jitter intensity of the mechanical load in the industrial power grid.

5. The industrial IoT data trusted transmission method based on edge computing according to claim 1, characterized in that, In S3, the step of inputting the multi-source time series acquisition data into a preset autoregressive prediction network layer and outputting the data drift prediction gradient of the multi-source time series acquisition data includes: The input nodes of the preset autoregressive prediction network layer receive the actual observed physical amplitude of the multi-source time series data at the current sampling time; The actual observed physical amplitude is weighted and summed with the historical predicted physical amplitude of the previous moment to output the target predicted physical amplitude of the next moment. Subtracting the actual observed physical amplitude from the predicted physical amplitude of the target physical amplitude yields the physical amplitude evolution deviation. The data drift prediction gradient is obtained by dividing the physical amplitude evolution deviation by a preset sampling time step constant.

6. The industrial IoT data trusted transmission method based on edge computing according to claim 1, characterized in that, In S4, the construction of the penalty coefficient dynamic optimization model, which inputs the industrial fault underreporting cost function and the physical link bandwidth congestion constraint boundary into the penalty coefficient dynamic optimization model, includes: Set a physical correlation sensitivity coefficient to characterize the penalty for threshold reduction and a network congestion dynamic pricing factor; The industrial failure underreporting cost function is generated by multiplying the basic downtime economic loss constant with a negative exponential function constructed from the physical correlation sensitivity coefficient. The burst throughput prediction value is obtained by dividing the preset network throughput adjustment weight by the physical correlation sensitivity coefficient. The physical link bandwidth congestion constraint boundary is generated by subtracting the physical network card's maximum communication bandwidth from the predicted burst throughput value. Multiplying the network congestion dynamic pricing factor by the physical link bandwidth congestion constraint boundary yields the Lagrange penalty term. The industrial fault underreporting cost function is added to the Lagrange penalty term to construct the Lagrange dual optimization equation, and the Lagrange dual optimization equation is used as the dynamic optimization model for the penalty coefficient.

7. The industrial IoT data trusted transmission method based on edge computing according to claim 6, characterized in that, In S4, the iterative calculation to output the optimal physical correlation sensitivity parameters includes: Obtain the first descent iteration step size and the second descent iteration step size; Based on the first descent iteration step size and the partial derivative expression of the penalty coefficient dynamic optimization model with respect to the physical association sensitivity coefficient, calculate and update the physical association sensitivity coefficient for the next iteration round; Based on the second descent iteration step size and the physical link bandwidth congestion constraint boundary, calculate and update the network congestion dynamic pricing factor for the next iteration round. Compare the absolute value of the difference between the physical correlation sensitivity coefficients of two adjacent iterations; When the absolute value of the difference is less than the built-in lower limit of the convergence error, the subgradient iteration calculation is stopped, and the physical correlation sensitivity coefficient in the converged state is output as the optimal physical correlation sensitivity parameter.

8. The industrial IoT data trusted transmission method based on edge computing according to claim 1, characterized in that, In S5, the step of fusing the optimal physical correlation sensitivity parameter, the local sliding variance, and the baseline static threshold to generate the optimal dynamic scheduling physical threshold includes: The optimal physical correlation sensitivity parameter is multiplied by the local sliding variance to output the power grid disturbance penalty correction amount; The baseline static threshold is read from the non-volatile memory, and the grid disturbance penalty correction is subtracted from the baseline static threshold to calculate the optimal dynamic scheduling physical threshold.

9. The industrial IoT data trusted transmission method based on edge computing according to claim 1, characterized in that, In S6, comparing the data drift prediction gradient with the optimal dynamic scheduling physical threshold, and generating preemptive transmission triggering signaling when the data drift prediction gradient is greater than or equal to the optimal dynamic scheduling physical threshold, includes: The data drift prediction gradient is input to the positive phase digital comparison pin of the hardware logic comparison circuit of the edge node device; The optimal dynamic scheduling physical threshold is input to the inverted digital comparison pin included in the hardware logic comparison circuit; When the input value of the positive digital comparator pin is greater than or equal to the input value of the negative digital comparator pin, a continuous high-level pulse is output to the network card interrupt controller on the motherboard of the edge node device through the system data bus, and the high-level pulse is used as the preemptive transmission trigger signaling.

10. The industrial IoT data trusted transmission method based on edge computing according to claim 1, characterized in that, In S7, the step of transferring the transparent transmission data payload containing the hardware-level physical timestamp and the multi-source time series acquisition data to the network interface card (NIC) transmission queue through the preemptive transmission trigger signaling, and performing a cross-level network transmission action, includes: The preemptive transmission trigger signaling is captured by the external interrupt pin of the network card interrupt controller, and the regular polling process of the underlying transmission control protocol being executed by the operating system is forcibly suspended. Activate the direct memory access channel and directly concatenate the hardware-level physical timestamp and the multi-source time series data in the system memory space to form the transparent transmission data payload that does not require secondary encapsulation by the application layer protocol. The transparent transmission data payload is forcibly and continuously copied to the hardware transmit buffer of the network card's media independent interface by the active direct memory access channel, and the hardware transmit buffer is used as the network card's transmit queue. Locate the byte offset position of the Differentiated Services Control field in the header structure layout, and rewrite the Differentiated Services Control field as the highest transmission priority identifier; By locking the exclusive right to use the transmission channel of the physical network interface device and bypassing the congestion control logic of the operating system's transport layer, the network data packet containing the transparent transmission data payload and the highest transmission priority identifier is directly sent to the external physical communication link, thus completing the cross-level network transmission action.

Images